HP Exemplar SPP-2000 addition to Community computing

The 64-processor Hewlett-Packard SPP-2000 (sioux) was originally acquired during the spring of 1997 as an experimental HPCC machine, and since early fall 1997, it has been used by a slowly growing number of "friendly" Community users. During FY1998, SCD moved the SPP-2000 from an experimental HPCC role into a more general, operational role by adding the SPP-2000 to the Community computational resources; sioux was formally transitioned to a Community computational resource in February 1998.

Making the HP SPP-2000 part of the Community computing resources at NCAR allowed willing users to take advantage of the new Distributed Shared Memory (DSM) architectures available in the U.S. supercomputer marketplace. The sioux system provides, in essence, a gateway to what appears to be the future of computational equipment available to NCAR users and an early opportunity to begin optimizing models and data analysis packages for scalar, cache-based, microprocessor systems.

SCD transitioned the sioux system into a "production" system, and our intentions were to provide, as much as possible, a user environment resembling that available on other Community supercomputers. This required the implementation of the Batch Dedicated Scheduler (BDS) that currently enforces wall-clock limits and turns batch queues on and off automatically to accommodate the varied needs of the small user community that has taken advantage of the system. The Unified Queuing System report provides more information.

Certainly, the SPP-UX operating system and product set differs substantially from the UNICOS or Irix environments with which Community users are familiar. Thus, users of the HP system were faced with learning a new environment, particularly for compilers, loaders, debuggers, and other application development utilities. SCD continues to strive, wherever possible, to make the HP environment similar to that provided on other compute servers at NCAR.

During FY1998, SCD performed the following activities to integrate the SPP-2000 into the NCAR Community computing environment:

Extensive system performance testing and tuning, including disk filesystem, memory partitioning, and kernel scheduling parameter tuning
Extensive NQS and checkpoint/restart testing and augmentation of vendor-supplied subsystems with enforcement of user-specified time and memory limits. Checkpoint/restart does not work reliably.
Implementation of a new queue structure and the Batch Dedicated Scheduler for DSMs
Ported and implemented many of the SCD-specific administrative procedures and tools. The NCAR scrubber is being ported in early FY1999.

The operating system provides no accounting capabilities, thus SCD has left the system unallocated but available to any Community users interested in obtaining an account and using the system. SCD has continued to closely monitor sioux and add, change, or tune resources as necessary to ensure the best possible system performance and job throughput. To date, this system has been relatively well utilized, and utilization has been increasing with time (approximately 70% utilized at end FY1998), but only by a small number of Community users.

1998 ASR Home

Back

SCD ASR Index

Next

SCD Home